Segmentation is the algorithm for tabletop/shelf segmentation. More...

#include <segmentation.h>

Public Member Functions
bool	Segment (std::vector< SurfaceObjects > *surfaces) const
	Segments the scene.
	Segmentation ()
	Default constructor.
void	set_cluster_distance (double cluster_distance)
	Sets the minimum distance between object clusters.
void	set_horizontal_tolerance_degrees (double degrees)
	Sets the tolerance for a surface to be considered horizontal.
void	set_indices (pcl::PointIndicesPtr indices)
	Sets the indices in the point cloud to use as input.
void	set_input_cloud (pcl::PointCloud< pcl::PointXYZRGB >::Ptr cloud)
	Sets the input cloud to use.
void	set_margin_above_surface (double margin)
	Sets a margin above the surface to start segmenting objects from.
void	set_max_cluster_size (int max_cluster_size)
	Sets the minimum cluster size.
void	set_max_point_distance (double max_point_distance)
	Sets the maximum distance of inlier points considered to be part of a surface.
void	set_min_cluster_size (int min_cluster_size)
	Sets the minimum cluster size.
void	set_min_surface_exploration_iteration (int min_surface_exploration_iteration)
	Sets the minimum number of iterations required when exploring surfaces in the input point cloud.
void	set_min_surface_size (int min_surface_size)
	Sets the minimum number of points a surface has to have.
Private Attributes
pcl::PointCloud < pcl::PointXYZRGB >::Ptr	cloud_
double	cluster_distance_
double	horizontal_tolerance_degrees_
pcl::PointIndicesPtr	indices_
double	margin_above_surface_
int	max_cluster_size_
double	max_point_distance_
int	min_cluster_size_
int	min_surface_exploration_iteration_
int	min_surface_size_

Detailed Description

Segmentation is the algorithm for tabletop/shelf segmentation.

This algorithm takes in a tabletop/shelf scene and segments it into surfaces with some number of objects above each surface. The objects are segmented using a Euclidean clustering algorithm. For each surface, The algorithm fits oriented bounding boxes around the surface and the objects. For each object, the z direction points "up," and the x direction points toward the shorter side of the oriented bounding box.

The algorithm assumes that the input scene is provided such that the positive z direction points "up."

Example usage:

   surface_perception::Segmentation seg;
   seg.set_input_cloud(pcl_cloud);
   seg.set_indices(point_indices);
   seg.set_horizontal_tolerance_degrees(10);
   seg.set_margin_above_surface(0.01);
   seg.set_cluster_distance(0.01);
   seg.set_min_cluster_size(10);
   seg.set_max_cluster_size(10000);
   seg.set_min_surface_size(10000);

   std::vector<SurfaceObjects> surface_objects;
   bool success = seg.Segment(&surface_objects);

Definition at line 42 of file segmentation.h.

Constructor & Destructor Documentation

surface_perception::Segmentation::Segmentation ( )

Default constructor.

Definition at line 30 of file segmentation.cpp.

Member Function Documentation

bool surface_perception::Segmentation::Segment ( std::vector< SurfaceObjects > * surfaces ) const

Segments the scene.

Parameters:

[out] surfaces The vector of SurfaceObjects to append to. This algorithm only supports tabletop segmentation at this time, so it will append one SurfaceObjects instance if successful.

Returns:: true if the segmentation was successful, false otherwise. Reasons why segmentation might fail include the surface not being found and no points were found above the surface.

Definition at line 77 of file segmentation.cpp.

void surface_perception::Segmentation::set_cluster_distance ( double cluster_distance )

Sets the minimum distance between object clusters.

The algorithm clusters nearby points and considers each cluster an object. This parameter is the threshold used to determine if two points are part of the same object or if they are part of different objects.

Parameters:

[in] cluster_distance The cluster distance, in the same units as the point cloud (usually meters).

Definition at line 57 of file segmentation.cpp.

void surface_perception::Segmentation::set_horizontal_tolerance_degrees ( double degrees )

Sets the tolerance for a surface to be considered horizontal.

Parameters:

[in] degrees The tolerance, in degrees, for a surface to be considered horizontal.

Definition at line 51 of file segmentation.cpp.

void surface_perception::Segmentation::set_indices ( pcl::PointIndicesPtr indices )

Sets the indices in the point cloud to use as input.

Definition at line 47 of file segmentation.cpp.

void surface_perception::Segmentation::set_input_cloud ( pcl::PointCloud< pcl::PointXYZRGB >::Ptr cloud )

Sets the input cloud to use.

The cloud should be provided such that the positive z direction points up.

Definition at line 42 of file segmentation.cpp.

void surface_perception::Segmentation::set_margin_above_surface ( double margin )

Sets a margin above the surface to start segmenting objects from.

The algorithm fits a plane to the surface. Some points above that plane will correspond to the surface, so the margin is used to ignore all points within a certain distance above the plane.

Parameters:

[in] margin The margin, in the same units as the point cloud (usually meters), above the fitted plane to consider as part of the surface.

Definition at line 54 of file segmentation.cpp.

void surface_perception::Segmentation::set_max_cluster_size ( int max_cluster_size )

Sets the minimum cluster size.

The algorithm clusters nearby points and considers each cluster an object. This parameter is used to filter out large parts of the scene that are above the surface but are not objects (e.g., a back wall). This parameter is sensitive to the density of the input point cloud.

Parameters:

[in] max_cluster_size The maximum number of points that can be in a cluster for the cluster to be considered an object.

Definition at line 63 of file segmentation.cpp.

void surface_perception::Segmentation::set_max_point_distance ( double max_point_distance )

Sets the maximum distance of inlier points considered to be part of a surface.

As the surface exploration algorithm explores surfaces in the given point cloud scene, the value of max_point_distance is used to compute the number of inlier points for a surface. The number of surface inliers is then used to determine the quality of the explored surfaces for the output.

Parameters:

[in] max_point_distance The maximum distance threshold of surface inlier points.

Definition at line 73 of file segmentation.cpp.

void surface_perception::Segmentation::set_min_cluster_size ( int min_cluster_size )

Sets the minimum cluster size.

The algorithm clusters nearby points and considers each cluster an object. This parameter is used as a filter to ignore noise. This parameter is sensitive to the density of the input point cloud.

Parameters:

[in] min_cluster_size The number of points that must be in a cluster for that cluster to be considered an object.

Definition at line 60 of file segmentation.cpp.

void surface_perception::Segmentation::set_min_surface_exploration_iteration ( int min_surface_exploration_iteration )

Sets the minimum number of iterations required when exploring surfaces in the input point cloud.

The surface exploration algorithm samples a height value for a horizontal at each iteration, and min_surface_exploration_iteration indicates the lower bound of number of iterations needed for the horizontal surface sampling process.

Parameters:

[in] min_surface_exploration_iteration The surface exploration algorithm needs to run at least the specified number of iteration during surface exploration.

Definition at line 69 of file segmentation.cpp.

void surface_perception::Segmentation::set_min_surface_size ( int min_surface_size )

Sets the minimum number of points a surface has to have.

As the algorithm only considers surfaces with more than required number of points as valid surfaces, the lower bound of surface size is used in order to ignore invalid surfaces

Parameters:

[in] min_surface_size The minimum requirement on number of points in a surface.

Definition at line 66 of file segmentation.cpp.